The 70% of AI projects that never reach production have something in common: they hit the authentication wall and never recover. Tradestack broke through it by leveraging Arcade’s capabilities.
When Vaibhav Pandey and his team at Tradestack set out to build an agentic back office for UK contractors, they faced a challenge that kills most agent projects: giving AI secure, reliable access to critical business systems. Their target customers (mid-market contractors juggling invoicing, estimates, and project management) needed more than a chatbot. They needed an agent that could actually do things, on the systems they already use.
"If the agent cannot take action on your behalf as a business owner, then the agent is practically not of much use," explains Pandey. For Tradestack's WhatsApp-based assistant to deliver real value, it needed to create estimates in job management systems, send quotes and invoices through Xero, and manage communications via email clients like Outlook and Gmail, all while maintaining enterprise-grade security.
The Authentication Wall That Stops 70% of Projects
Most teams discover the hard way that building production agents requires expertise that simply doesn't exist in the market. You need the intersection of AI, identity systems, and distributed infrastructure knowledge—a combination so rare it's effectively a null set in hiring.
Tradestack initially explored existing solutions but found pre-built tools offered "less control than we wanted." As Pandey notes, "We felt that's not the ideal experience where we have pre-cooked tools and we are not able to influence the tool design."
The authentication challenge goes deeper than just OAuth flows. When your AI touches financial data, customer communications, and business-critical systems, trust becomes existential. Pandey compares it to "swiping your card four times" at checkout—even if you know you'll only be charged once, the friction creates doubt.
From Concept to Production: The Two-Week Sprint
After a recommendation from Harrison Chase at LangChain, Tradestack evaluated Arcade.dev. The difference was immediate: clean OAuth implementation, control over tool design, and (crucially) a team that understood both AI behavior and enterprise security.
"Nate and Mateo were very helpful in fixing issues very quickly," Pandey recalls. "We went from zero to two toolkits in production—Outlook and Xero—within a few weeks."
This wasn't just about speed. Tradestack needed production-grade infrastructure from day one:
- Enterprise Authorization: Not just "who is this user?" but "should this user's agent be allowed to create this invoice right now?"
- Granular Permissions: Agents that can read emails but not delete them, respond only to customer enquiries, create draft invoices but not approve payments
- Customer Trust Signals: Using Arcade's auth screen helped establish credibility—"as a business owner, you don’t want your account to be accessed through an untrusted app" as Pandey puts it
Building for the Real World, Not Demos
Tradestack's production deployment revealed insights that only emerge when agents face actual customers:
1. Clean OAuth is Non-Negotiable "Having a very clean OAuth experience is very important for customer trust," Pandey emphasizes. In the UK construction industry, where business owners or their IT teams must approve email access, and financial data is sacred, any authentication friction becomes a deal-breaker.
2. Quality Over Quantity in Integrations Instead of chasing sheer numbers, Tradestack has deliberately focused on building 4–5 deeply integrated, high-quality toolkits that truly matter to customers’ workflows. “We don’t see ourselves adding 20 toolkits for the sake of it. What we do see is carefully investing in four or five integrations that are indispensable to our users. That’s the essence of what we’ve learned, and why Arcade feels like such a strong platform for building agent capabilities. Unlike many platforms that highlight thousands of pre-built integrations, the Arcade team seems to uniquely grasp what they call ‘machine-experience’ principles. To us, our system needs real competence rather than signalling surface-level capability. The moment you try to reliably automate even a single task, you realize just how many nuances and edge cases must be handled with care. In that process, tool design, what we consider a subset of context engineering, emerges as a decisive performance lever. We would always prefer four integrations that work seamlessly and intelligently over a hundred that are half-baked.”
3. Human-in-the-Loop for High-Stakes Actions: HITL can take many forms, but Tradestack found the most valuable approach is when the agent knows when to ask for approval. Take a simple workflow: converting a client enquiry into a work order once a quote is accepted. Most of the time, no human check is necessary. But when something looks off, say, the client-approved price doesn’t match the original quote, the scope of work has shifted, or there are signs of a duplicate, the system can pause and flag it for review. Even for tasks that go through without intervention, users can still provide feedback afterward, and the agent incorporates that learning for the future. Over time, this feedback loop helps the system recognize which tool calls or scenarios are routine, and which are context-sensitive enough to require human judgment.
The Architectural Advantage
What enabled Tradestack to move so fast while maintaining enterprise standards? They didn't have to solve the fundamental architectural mismatches that plague agent development:
- LLM-to-API Translation: Arcade's tools speak "LLM intentions" rather than forcing agents to navigate complex API structures
- Post-Prompt Authorization: Security decisions made after understanding user intent, not before
- Production Infrastructure: Deployment, monitoring, and scaling handled by the platform
As Pandey explains, Tradestack viewed Arcade as their "internal MCP before MCP existed"—a central platform for creating, deploying, and managing all agent capabilities.
Results: From WhatsApp and Email to Business Action
Today, Tradestack's solution helps UK contractors delegate critical back office tasks through simple WhatsApp messages and access to Email inbox:
- "Prepare a cost estimate for the loft conversion in Islington and send it to Michael for review"
- "Generate invoices for last week's completed jobs and submit for payment"
- "Check my email and ensure all supplier bills are added to Xero"
Each request triggers secure, authorized actions across multiple systems—all happening reliably in production, not just in carefully controlled demos. The ability for business owners to maintain clear control over projects and finances, while knowing the tasks they’ve delegated to Tradestack are executed reliably and on time, unlocks significant productivity gains across a sector that has long been underserved.
The Path Forward: Scaling Intelligence, Not Infrastructure
With authentication and infrastructure solved, Tradestack can focus on what actually matters: building better intelligence for their customers. Their roadmap includes:
- Context and Long-Term Memory: Tracking project conversations across multiple platforms
- Trajectory Optimization: Caching consistent workflows (like supplier bill uploads) for faster, more reliable execution
- Scalable Evaluation: Moving beyond manual testing to systematic performance tracking across different customers and use cases
The Lesson: Choose Your Battles
"Authorization is rocket science," as one enterprise architect told us. Tradestack succeeded by recognizing this truth early and choosing a platform that had already solved it.
While 70% of AI agent projects fail trying to rebuild authentication infrastructure, Tradestack went from concept to production in two weeks. They didn't just move fast—they built production-grade agents that handle financial data, customer communications, and business operations with the security and reliability their enterprise customers demand.
The difference? They focused on building intelligence for UK contractors, not wrestling with OAuth flows and token management. Sometimes the smartest architectural decision is knowing what not to build.
Tradestack's AI assistant is transforming how UK construction businesses manage their operations. Learn how you can build production-grade agents without the infrastructure headaches at arcade.dev.
